Filters For Low-Noise Measurement Preamplifier

[bryman79]

Originally Posted by jcx

the potential problem is if the noise is large out of band then the input amp developing all of the gain in one step may clip and cause nonlinear mixing, both giving incorrect measurements if followed by unity gain filters

Thanks for pointing this out.

Can you direct me to information on nonlinear mixing in this application? I looked online and found a lot of IEEE papers and optics-related articles, but nothing that I could access related to noise or audio amplifiers.

From what I could find, an in-band frequency can be created by one or more out-of-band frequencies in three instances: summation, difference, and harmonics. Anyone know how the amplitude of the in-band frequency is calculated in each instance with respect to gain?

-Bryan

[Samuel Groner]

It's been ~1K all along.

My bad, apologies. I didn't realise that the schematic omitts the first stage.

Can you direct me to information on nonlinear mixing in this application?

I don't think there is any specific literature on this, nor the need for any. The rule is simple: no amplifier may clip during the measurement, or the result could be invalid. The problem with the filters following the amplifier is that they may remove the obvious clipping artefacts. It is wise to add a BNC output before the filter such that you can monitor the signal there as well.

There is an inherent trade-off between headroom (or overload margin or insensitivity to out-of-band signals) and low noise. For lowest noise you want to amplify the signal first, then filter. That way the typically rather noisy filter stages are less significant noise contributors as the signal has been lifted above their noise floor. For best headroom you want to filter first (to minimize signal energy and remove out-of-band content), then amplify.

There exists an infinite number of compromises in between these two extremes--first amplify a bit, then filter, then again an amplifier. There are even commercial amplifiers where you can switch between two modes--lowest noise or best headroom.

In my preamp design you asked about in some e-mail I kept first stage gain low to get very good headroom. At the cost of some noise performance.

It get's back to what I said above: you need to know what you want to do with this preamp. If you're measuring discrete high-performance linear power supplies there is likely not much out-of-band energy, but you want lowest noise to not influence the measurement. If you're investigating standard switching regulators noise contribution from the preamp is hardly a difficult issue; however, there's considerable high-frequency energy around which needs to be removed before significant amplification can take place.

Samuel

[bryman79]

Originally Posted by Samuel Groner

In my preamp design you asked about in some e-mail I kept first stage gain low to get very good headroom. At the cost of some noise performance.

Speaking of your preamp, what is the purpose of R1 (10M) on the input to ground?

Originally Posted by Samuel Groner

If you're investigating standard switching regulators noise contribution from the preamp is hardly a difficult issue; however, there's considerable high-frequency energy around which needs to be removed before significant amplification can take place.

Although I had no interest in measuring the noise of a switching PSU a month ago, that has changed. Can you recommend a lowpass filter that could be used on the input with minimal impact on noise performance? I tried an RL filter on the input (similar to the Wenzel amp), which worked well for the AD797 60dB stage but not so well for the designs with the FET input stages.

Thanks,
Bryan

[Samuel Groner]

What is the purpose of R1 (10M) on the input to ground?

It discharges the input coupling capacitor if the preamp is disconnected from a still powered supply. The stored energy in the capacitor could otherwise damage either the next device you connect the preamp to, or the preamp itself.

Can you recommend a lowpass filter that could be used on the input with minimal impact on noise performance?

I don't think you need to worry about noise impact with a switcher. A 1k/1nF RC section should do the trick, usually (and work with either BJT or JFET front-ends).

Samuel